neoctl 0.2.58 → 0.2.60

package/dist/types/events.d.ts

@@ -30,6 +30,8 @@ export interface PromptCacheDiagnostics {
 export interface ContextMetrics {
     model?: string;
     estimatedInputTokens: number;
+    /** Prefer this unified estimate/count over generation usage for context display. */
+    inputTokenSource?: "estimate";
     estimatedChars: number;
     messageCount: number;
     toolCount: number;

package/dist/web/html.js

@@ -405,7 +405,7 @@ function renderStatus() {
   const displayPhase = minimumDisplayPhase(s.phase || 'ready');
   const phase = phaseLabel(displayPhase);
   const ctx = contextParts(s.metrics);
-  const inputTokens = compactNumber(animatedNumber('input', (s.usage && s.usage.inputTokens) ?? (s.metrics && s.metrics.estimatedInputTokens)));
+  const inputTokens = compactNumber(animatedNumber('input', s.metrics?.inputTokenSource ? s.metrics.estimatedInputTokens : (s.usage && s.usage.inputTokens) ?? (s.metrics && s.metrics.estimatedInputTokens)));
   const outputTokens = compactNumber(animatedNumber('output', (s.usage && s.usage.outputTokens) ?? s.streamedOutputTokens));
   const model = truncateMiddle((s.metrics && s.metrics.model) || 'model?', window.innerWidth > 900 ? 26 : 14);
   const phaseActive = isActivePhase(displayPhase);

package/dist/web/index.js

@@ -3,8 +3,8 @@ import { trackSettingsGateway } from "../model/settings-sync.js";
 import { observeModelSettings } from "../model/settings-sync.js";
 import { modelProtocols } from "../model/context-window.js";
 import { isReasoningEffort } from "../model/model-gateway.js";
-import { ModelSettingsStore, compositionModelConfig } from "../model/settings-store.js";
-import { registerCompositionTools } from "../model/composition-runtime.js";
+import { ModelSettingsStore, routeModel } from "../model/settings-store.js";
+import { registerCompositionTools, createCompositionGateway } from "../model/composition-runtime.js";
 import { handleModelSettingsRequest } from "./model-settings-api.js";
 import { paginateSessions, sessionPageOptions } from "../session/session-pagination.js";
 import { applyToolInputChanges } from "../model/tool-input-stream.js";
@@ -158,9 +158,10 @@ export async function createWebRuntime(options = {}) {
     const settingsStore = new ModelSettingsStore();
     const appliedSettings = settingsStore.read()?.applied;
     const settingsRevisions = new WeakMap();
-    const modelConfig = appliedSettings ? compositionModelConfig(appliedSettings.composition, appliedSettings.connections) : readModelProviderConfig(process.env);
+    const appliedGateway = appliedSettings ? createCompositionGateway(appliedSettings.composition, appliedSettings.connections) : undefined;
+    const modelConfig = appliedGateway ? appliedGateway.config : readModelProviderConfig(process.env);
     const communicationLogger = new CommunicationLogger();
-    const modelGateway = trackSettingsGateway(new LoggingModelGateway(createModelGatewayFromConfig(modelConfig), communicationLogger), appliedSettings?.revision, appliedSettings ? [appliedSettings.composition.main.api, appliedSettings.composition.image.api, appliedSettings.composition.search.api] : []);
+    const modelGateway = trackSettingsGateway(new LoggingModelGateway(appliedGateway ? appliedGateway.gateway : createModelGatewayFromConfig(modelConfig), communicationLogger), appliedSettings?.revision, appliedGateway ? appliedGateway.heldConnections : []);
     const taskStore = new TaskStore();
     const pluginCatalog = normalizeWebRuntimePlugins(options.externalPlugins);
     const pluginOverrides = normalizePluginOverrides(options.sessionPluginOverrides, pluginCatalog);
@@ -252,11 +253,12 @@ export async function createWebRuntime(options = {}) {
             const revision = applied.revision;
             if (settingsRevisions.get(engine) === revision)
                 return;
-            const config = compositionModelConfig(applied.composition, applied.connections);
-            const gateway = trackSettingsGateway(new LoggingModelGateway(createModelGatewayFromConfig(config), communicationLogger), revision, [applied.composition.main.api, applied.composition.image.api, applied.composition.search.api]);
+            const routed = createCompositionGateway(applied.composition, applied.connections), config = routed.config;
+            const gateway = trackSettingsGateway(new LoggingModelGateway(routed.gateway, communicationLogger), revision, routed.heldConnections);
             engine.setModelProvider({ modelGateway: gateway, model: config?.model, reasoning: config?.defaultReasoning });
             if (config)
                 effectiveModelProviders.set(engine, config.provider);
+            effectiveModelRoutes.set(engine, applied);
             registerCompositionTools(tools, applied.composition, applied.connections);
             for (const name of ["web_search", "image_create"]) {
                 const tool = tools.get(name);
@@ -1542,10 +1544,17 @@ export class WebRepl {
         }
         const engine = this.runtime.engine;
         const nextModel = model.trim();
-        const provider = effectiveModelProviders.get(engine) ?? readModelProviderConfig()?.provider;
-        const metadata = (engine.getActiveModelCatalog?.() ?? loadModelCatalog()).models.find(m => m.modelIds.includes(nextModel));
-        if (metadata && provider && !modelProtocols(metadata).includes(provider))
-            return actionFailure("MODEL_INVALID", "模型不支持当前主连接接口");
+        const routes = appliedRoutes(engine);
+        if (routes) {
+            if (!routeModel(nextModel, routes.composition, routes.connections))
+                return actionFailure("MODEL_INVALID", "没有已启用的 API 提供此模型");
+        }
+        else {
+            const provider = effectiveModelProviders.get(engine) ?? readModelProviderConfig()?.provider;
+            const metadata = (engine.getActiveModelCatalog?.() ?? loadModelCatalog()).models.find(m => m.modelIds.includes(nextModel));
+            if (metadata && provider && !modelProtocols(metadata).includes(provider))
+                return actionFailure("MODEL_INVALID", "模型不支持当前主连接接口");
+        }
         const current = { ...engine.getModelSettings(), ...engine.getPendingSessionSettings() };
         const argument = reasoning;
         const validationError = validateModelReasoningArgument(nextModel, argument);
@@ -3211,10 +3220,19 @@ function sanitizeWebAttachments(value) {
     });
 }
 const effectiveModelProviders = new WeakMap();
+const effectiveModelRoutes = new WeakMap();
+/** Applied priority list for this engine (or the stored one before its first turn); undefined for legacy/env setups. */
+function appliedRoutes(engine) {
+    const applied = effectiveModelRoutes.get(engine) ?? new ModelSettingsStore().read()?.applied;
+    return applied?.composition.order ? applied : undefined;
+}
 function webCatalog(runtime) {
     const catalog = runtime.engine.getActiveModelCatalog?.() ?? loadModelCatalog();
     const provider = effectiveModelProviders.get(runtime.engine) ?? readModelProviderConfig()?.provider;
-    const models = catalog.models.filter(model => !provider || modelProtocols(model).includes(provider));
+    const routes = appliedRoutes(runtime.engine);
+    const models = routes
+        ? catalog.models.filter(model => routeModel(model.id, routes.composition, routes.connections))
+        : catalog.models.filter(model => !provider || modelProtocols(model).includes(provider));
     const modelIds = [...new Set(models.flatMap(model => model.modelIds.length ? model.modelIds : [model.id]))].sort((left, right) => left.localeCompare(right, "en", { numeric: true }));
     return {
         commands: replCommandDefinitions,